A Web-Based Information System for Ornamental Rock Classification and Similarity Search Using Siamese Networks

Abstract

Research Context: The time-consuming, subjective, and error-prone process for identification of ornamental rocks, which is traditionally based on visual analysis and expert knowledge. This task is vital for competitiveness in construction, mining, and geology. Advances in computer vision create opportunities to automate such tasks and provide aid for professionals. Scientific/Practical Problem: The ornamental rock industry faces challenges in accurately and efficiently classifying rocks and retrieving similar units for high-demand users. The current manual effort is imprecise and slow, highlighting the need for AI solutions to improve the process. Proposed Solution and/or Analysis: This work proposes a web-based information system that uses Siamese Networks to aid analysts and general users in finding the class of a given ornamental rock, and its most similar pair of images in a dataset. Related IS Theory: Task-Technology Fit (TTF) guides our work. The proposed system is designed to allow smoother and quicker task completion, reducing time and effort. In addition, users can more easily and effectively achieve their desired outcomes. The smoother task execution can lead to lower costs associated with task performance. Research Method: A system is proposed using Siamese Networks to perform classification and similarity recognition of ornamental rocks. Summary of Results: The system successfully finds the class of the given image, as well as the image pair that most resembles it. The Siamese network comparison technique also allows for the correct identification of classes not used in training, but existing in the user’s database. Contributions and Impact to IS area: This work bridges advanced vision models and decision support in ornamental rocks identification, transforming a neural network model into a reliable, evidence-based system and task-aligned evaluation.